Histograms | The CUDA Handbook

← All benchmarks · Chapter 16

Some of these measurements depend on the host as much as on the GPU: launch overhead is CPU dispatch cost, transfer rates are bounded by the host link and the host's memory system, and the CPU baselines run entirely on the host. Results for one GPU are reported separately per host architecture whenever it has been measured on more than one.

![histogram.per-grid: Mpix/s by values count (log-log; book-era curves join when baselines land)](/benchmarks/media/histogram-per-grid.svg) ![histogram.per-block: Mpix/s by values count](/benchmarks/media/histogram-per-block.svg)

histogram.cpu.multi (Gpix/s) -- histogram/histogram.cu: multithreaded CPU reference

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	7.4	8.1	9.6	6.7	10.8	7.6	5.5	3.8	9.4	2

histogram.cpu.single (Gpix/s) -- histogram/histogram.cu: single-threaded CPU reference

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	4.2	4.1	4.1	4.1	4.1	3.9	2.9	1.7	4.8	2

histogram.npp (Gpix/s) -- histogram/histogram.cu: NVIDIA Performance Primitives nppiHistogramEven

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	280.6	281.9	285.3	283.5	246.4	198.0	155.9	138.7	291.6	2

histogram.per-block (Gpix/s) -- histogram/histogram.cu: per-block privatized histograms in shared memory, atomics

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	107.7	107.9	108.1	107.8	108.0	107.6	107.7	107.9	107.7	2

histogram.per-block-4x (Gpix/s) -- histogram/histogram.cu: per-block, 4 pixels per thread

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	278.7	277.8	279.9	280.8	280.7	278.7	279.5	276.9	280.8	2

histogram.per-block-4x-offset (Gpix/s) -- histogram/histogram.cu: per-block, 4x unrolled, warp-offset

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	275.8	279.7	279.7	278.6	281.4	278.6	279.3	279.6	282.0	2

histogram.per-block-offset (Gpix/s) -- histogram/histogram.cu: per-block with per-warp offset to spread contention

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	106.6	107.0	107.0	107.1	107.3	107.2	107.1	107.0	107.0	2

histogram.per-block-reduce (Gpix/s) -- histogram/histogram.cu: per-block histograms merged by reduction

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	161.4	162.6	161.3	162.0	162.3	162.2	162.1	163.6	162.2	2

histogram.per-block-reduce-offset (Gpix/s) -- histogram/histogram.cu: per-block reduction merge, warp-offset

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	160.8	160.7	161.7	162.3	162.7	161.1	161.0	162.0	162.6	2

histogram.per-grid (Gpix/s) -- histogram/histogram.cu: one histogram in global memory, atomics fired per-pixel

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	3.5	4.8	3.2	4.2	4.5	4.4	2.8	1.7	1.7	2

histogram.per-thread-4x32 (Gpix/s) -- histogram/histogram.cu: per-thread histograms, 32-count cells, 4x unrolled

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	45.8	45.9	45.9	46.0	46.0	45.8	45.7	45.8	45.9	2

histogram.per-thread-4x32-periodic-merge (Gpix/s) -- histogram/histogram.cu: per-thread 4x32 with periodic merge

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	63.6	64.2	64.5	64.3	64.0	63.9	63.9	64.8	64.0	2

histogram.per-thread-4x64 (Gpix/s) -- histogram/histogram.cu: per-thread 64-count, 4x unrolled

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	65.4	65.8	66.2	65.9	66.3	66.0	66.5	66.5	66.6	2

histogram.per-thread-4x64-periodic-merge (Gpix/s) -- histogram/histogram.cu: per-thread 4x64 with periodic merge

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	71.2	71.1	71.1	70.8	70.7	70.9	71.3	70.8	71.6	2

histogram.per-thread-64 (Gpix/s) -- histogram/histogram.cu: per-thread histograms, 64-count cells

GPU	256	128	64	32	16	8	4	2	1	n
Ampere
GeForce RTX 3060	25.7	25.7	25.7	25.7	25.7	25.7	25.7	25.7	25.7	2

Submissions

rtx-3060 · nwilt-2026-07-12-2.json -- Nicholas Wilt · driver 580.95.05 · CUDA 12.9
rtx-3060 · nwilt-2026-08-27.json -- Nicholas Wilt · driver 580.95.05 · CUDA 13.0